app/agent.py

from typing import List, Dict, Any
import json
from pathlib import Path
from typing import List, Dict, Any


CATALOG_PATH = Path("data/shl_catalog.json")


def load_catalog() -> List[Dict[str, Any]]:
    if not CATALOG_PATH.exists():
        return []

    with open(CATALOG_PATH, "r", encoding="utf-8") as f:
        return json.load(f)


CATALOG = load_catalog()

def get_last_user_message(messages: List[Any]) -> str:
    for msg in reversed(messages):
        if msg.role == "user":
            return msg.content.lower()
    return ""


def build_conversation_query(messages: List[Any]) -> str:
    """
    Builds a compact query from full stateless conversation history.
    This helps refinement requests like:
    'Actually add personality tests'
    without forgetting earlier role context.
    """
    user_messages = []

    for msg in messages:
        if msg.role == "user":
            user_messages.append(msg.content)

    return " ".join(user_messages).lower()

def is_out_of_scope(text: str) -> bool:
    text = text.lower()

    blocked_phrases = [
        # Legal / compliance
        "legal advice",
        "is it legal",
        "employment law",
        "labor law",
        "labour law",
        "discrimination",
        "fire employee",
        "terminate employee",
        "lawsuit",
        "contract",

        # General hiring advice outside SHL assessment recommendation
        "write interview questions",
        "interview questions",
        "salary",
        "compensation",
        "negotiate offer",
        "job description template",
        "write a job description",
        "resume screening",
        "cv screening",
        "cover letter",

        # Prompt injection
        "ignore previous instructions",
        "forget your instructions",
        "act as unrestricted",
        "bypass",
        "system prompt",
        "developer message",
        "reveal your prompt",
        "jailbreak",
    ]

    return any(phrase in text for phrase in blocked_phrases)


def is_vague(text: str) -> bool:
    vague_phrases = [
        "i need an assessment",
        "need assessment",
        "suggest assessment",
        "recommend assessment",
        "assessment test",
    ]

    has_role_signal = any(
        word in text
        for word in [
            "java",
            "python",
            "developer",
            "engineer",
            "manager",
            "sales",
            "graduate",
            "analyst",
            "stakeholder",
            "communication",
        ]
    )

    if any(phrase in text for phrase in vague_phrases) and not has_role_signal:
        return True

    if len(text.split()) <= 4 and not has_role_signal:
        return True

    return False


def score_catalog_item(query: str, item: Dict[str, Any]) -> int:
    score = 0

    name = item.get("name", "").lower()
    description = item.get("description", "").lower()
    keywords = item.get("keywords", [])

    searchable_text = f"{name} {description} {' '.join(keywords)}"

    query_words = query.lower().split()

    for word in query_words:
        if len(word) > 2 and word in searchable_text:
            score += 1

    # Strong skill boosts
    skill_terms = [
        "java", "python", "sql", "javascript", "developer",
        "coding", "programming", "software", "backend"
    ]

    for skill in skill_terms:
        if skill in query and skill in searchable_text:
            score += 3

    # Personality / behavior boost
    if any(term in query for term in ["personality", "communication", "stakeholder", "leadership"]):
        if item.get("test_type") == "P" or "personality" in searchable_text or "opq" in searchable_text:
            score += 3

    # Cognitive / aptitude boost
    if any(term in query for term in ["cognitive", "aptitude", "reasoning", "ability"]):
        if item.get("test_type") == "A" or "ability" in searchable_text:
            score += 3

    return score

def recommend(query: str) -> List[Dict[str, str]]:
    scored_items = []

    for item in CATALOG:
        score = score_catalog_item(query, item)
        if score > 0:
            scored_items.append((score, item))

    scored_items.sort(key=lambda x: x[0], reverse=True)

    recommendations = []
    for _, item in scored_items[:10]:
        recommendations.append(
            {
                "name": item.get('name',''),
                "url": item.get('url',''),
                "test_type": item.get('test_type','unknown'),
            }
        )

    return recommendations

def is_compare_query(text: str) -> bool:
    compare_terms = [
        "difference between",
        "compare",
        "vs",
        "versus",
        "different from",
        "which is better",
    ]

    text = text.lower()
    return any(term in text for term in compare_terms)


def find_matching_assessments(text: str, limit: int = 5) -> List[Dict[str, Any]]:
    text = text.lower()
    matches = []

    for item in CATALOG:
        name = item.get("name", "").lower()
        description = item.get("description", "").lower()
        keywords = " ".join(item.get("keywords", [])).lower()

        searchable_text = f"{name} {description} {keywords}"

        score = 0

        # Direct name match
        for token in text.split():
            if len(token) > 2 and token in name:
                score += 3
            elif len(token) > 2 and token in searchable_text:
                score += 1

        if score > 0:
            matches.append((score, item))

    matches.sort(key=lambda x: x[0], reverse=True)

    return [item for _, item in matches[:limit]]





def compare_assessments(query: str) -> Dict[str, Any]:
    matches = find_matching_assessments(query, limit=4)

    if len(matches) < 2:
        return {
            "reply": "I can compare SHL assessments, but I need two assessment names from the catalog. Please mention both assessments you want to compare.",
            "recommendations": [],
            "end_of_conversation": False,
        }

    a = matches[0]
    b = matches[1]

    a_name = a.get("name", "Assessment 1")
    b_name = b.get("name", "Assessment 2")

    a_type = a.get("test_type", "Unknown")
    b_type = b.get("test_type", "Unknown")

    a_desc = a.get("description", "")
    b_desc = b.get("description", "")

    reply = (
        f"Here is a catalog-grounded comparison:\n\n"
        f"{a_name} focuses on: {a_desc[:300] if a_desc else 'description not available in catalog data'}\n"
        f"Test type: {a_type}\n\n"
        f"{b_name} focuses on: {b_desc[:300] if b_desc else 'description not available in catalog data'}\n"
        f"Test type: {b_type}\n\n"
        f"Use {a_name} when the role requirement matches its catalog description. "
        f"Use {b_name} when the hiring need is closer to its catalog description."
    )

    return {
        "reply": reply,
        "recommendations": [
            {
                "name": a.get("name", ""),
                "url": a.get("url", ""),
                "test_type": a.get("test_type", "Unknown"),
            },
            {
                "name": b.get("name", ""),
                "url": b.get("url", ""),
                "test_type": b.get("test_type", "Unknown"),
            },
        ],
        "end_of_conversation": False,
    }
    

def run_agent(messages: List[Any]) -> Dict[str, Any]:
    query = build_conversation_query(messages)
    last_user_query = get_last_user_message(messages)

    if not query:
        return {
            "reply": "Please share the role or skills you want to assess using SHL assessments.",
            "recommendations": [],
            "end_of_conversation": False,
        }

    if is_out_of_scope(query):
        return {
            "reply": "I can only help with SHL assessment recommendations, comparisons, and refinements based on the SHL catalog.",
            "recommendations": [],
            "end_of_conversation": False,
        }

    if is_vague(query):
        return {
            "reply": "Sure. Which role, skill area, and seniority level are you hiring for?",
            "recommendations": [],
            "end_of_conversation": False,
        }

    # Important safety check
    # If real SHL catalog JSON is missing or not loaded, don't crash and don't hallucinate.
    if not CATALOG:
        return {
            "reply": "The SHL catalog is not loaded yet. Please try again after the catalog data is available.",
            "recommendations": [],
            "end_of_conversation": False,
        }

    if is_compare_query(last_user_query):
        return compare_assessments(last_user_query)

    recommendations = recommend(query)

    if recommendations:
        return {
            "reply": f"Based on the role details, here are {len(recommendations)} SHL assessments that may fit this hiring need.",
            "recommendations": recommendations,
            "end_of_conversation": False,
        }

    return {
        "reply": "Could you share the target role, required skills, and seniority level so I can recommend relevant SHL assessments?",
        "recommendations": [],
        "end_of_conversation": False,
    }